Skip to content

SECURITY.md

Latest commit

 

History

History
67 lines (48 loc) · 2.46 KB

SECURITY.md

File metadata and controls

67 lines (48 loc) · 2.46 KB
Raw

Security Policy

LAD-A2A is a discovery-and-trust bootstrap protocol whose threat model assumes local networks are hostile by default (see spec §1.4). Security reports are therefore taken seriously and prioritized.

Supported Versions

The protocol and reference implementation are pre-1.0 and evolving. Security fixes are applied to the main branch and the latest tagged draft.

Component Version Supported
Specification 0.1.0-draft ✅
Reference implementation 0.1.x ✅
Older drafts / tags < 0.1.0 ❌

Reporting a Vulnerability

Please do not open a public GitHub issue for security vulnerabilities.

Report privately through one of:

  1. GitHub Security Advisories (preferred) — open a draft advisory at https://github.com/franzvill/lad/security/advisories/new.
  2. Email — send details to the maintainer listed on the repository profile, with LAD-A2A SECURITY in the subject line.

Please include:

  • A description of the issue and the component affected (spec, server, client, signing, demo).
  • Steps to reproduce or a proof of concept.
  • The impact you foresee (e.g. spoofed agent accepted, unverified card marked verified, MITM downgrade).

What to expect

  • Acknowledgement within 5 business days.
  • An assessment and, where accepted, a remediation plan with a target timeline.
  • Public disclosure coordinated with you once a fix is available, with credit if desired.

Scope

In scope:

  • Trust-model soundness in the reference client/server (verification, consent, signing) — e.g. an unverified agent being reported as verified.
  • Discovery-path weaknesses (mDNS spoofing, well-known/captive-portal injection, TLS downgrade) that the spec claims to defend against.
  • Signing/verification flaws in common/signing.py.

Out of scope:

  • Vulnerabilities in upstream dependencies (report those upstream; we will bump the pin once a fix ships).
  • The demo application's LLM backend or API keys — the demo is illustrative and not intended for production use.
  • Denial of service from an on-path attacker who already controls the local network beyond what the threat model covers.

Security Expectations for Deployers

The reference implementation ships with safe defaults (TLS verification on, unverified agents denied by the default consent callback). Before deploying, follow PRODUCTION-CHECKLIST.md.